1 / 16

Job Priorities WG

Job Priorities WG. Dietrich Liko IT/PSS. Acknowledgments. I am reporting on behalf of the Job Priority WG For the work in the last weeks I want to thank several persons … Simone Camapna , Andrea Sciaba , Jeff Templon , Di Quing , Gergely Debreczeni

kostya
Download Presentation

Job Priorities WG

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Job Priorities WG Dietrich Liko IT/PSS

  2. Acknowledgments • I am reporting on behalf of the Job Priority WG • For the work in the last weeks I want to thank several persons … • Simone Camapna, Andrea Sciaba, Jeff Templon, Di Quing, GergelyDebreczeni • And others who participated in the discussion …

  3. Introduction • We have presented last year a model to implement a simple strategy for Job Priorities based on the VOMS FAQN of the user certificate. • While the principle seems straight forward, in the deployment several issues were found that have prevented a successful use of this model • A strategy has been developed to address these problems

  4. Reminder: VO requirements • Driven by ATLAS and CMS • Few groups • Example: ATLAS • Production, Users and Software managers • Simple rules ( e.g. Production 80%, Priority for Software managers etc.) • Aim • Gain fast experience with a simple model • Keep in mind that in the future a more complex model might be needed

  5. JPWG Model • Fair share between users on the batch system • Several shares in the batch system for different groups. VOMS FAQN is used to transmit the group • Publication of the use of the share in the Information System using the VOView mechanism • gLite WMS uses this information to select the best site for a job

  6. Complications • First time use of VOMS certificate • Interaction with the Data Management • Example LFC • Use of correct ACLs • Introduction of secondary groups • Difference in the matching mechanism for VOMS FAQN used by the CE and the WMS • CE: LCMAPS mechanism • WMS: VOView mechanism • Fixing the last problem requires not only software modification, but is also an deployment issue • Could not be done on the fly

  7. REDISCUSSION • What to do at this point • Asked for input from the experiments • Keep it really simple • Have a minimal version operational fast • ATLAS production needs a solution as fast as possible

  8. SEVERAL Measures • Very short time • Remove all current configurations • Short time • Implement an ad-hoc solution • Medium time • Look into other solution • New TCG input • Concentrate on the short time • For all other aspects the WG has been put to sleep • Waiting for the input from the Authorisation (Christoph)

  9. AN EXAMPLE • Three shares configured in the batch system and published using the VOView mechanism • Special share for Production • Access Rule: VOMS:/atlas/Role=production • Special share for Software Manager • Access Rule: VOMS:/atlas/Role=software • Catch all share for everybody else • Access Rule: VO:atlas

  10. THE PROBLEM • For the WMS all shares enter like independent CE • WMS does not know that they all belong together and LCMAPS will send the job to the right share • Mismatch of the mechanism between the CE and the WMS

  11. THE AD-HOC solution • Add DENY tags to the catch all share • Production Share • VOMS:/atlas/Role=production • Software Share • VOMS:/atlas/Role=software • Catch all Share • VO:atlas • DENY: /atlas/Role=production • DENY: /atlas/Role=software

  12. CRITISISM • Evidently its not nice to introduce such an ad-hoc tag • We do not know how long we have to live with it • But it is the only reasonable strategy • The better solution requires a significant modification of the CE matching implementation • That cannot be done on a short time scale • The timescale would be very uncertain • An document describing this solution and other aspects can be found on WG Wiki • Latest version with some updates

  13. Practical ISSUES • Modification the Dynamic Information provider • Modification of yaim scripts to generate complex configuration scripts • Careful testing of the full configuration

  14. How are we following up • Tight loop between the yaim developers, grid deployment and the WG • Thanks for their effort! • Test on the certification test bed are progressing • Several persons involved • Holiday, conferences etc … • If we are satisfied with the state on the certification testbed, we will move to preproduction

  15. FINAL issue • Experiments are interested in a reasonable default batch system configuration • For example: • Production share: ATLAS 80 %, CMS 50 % • If the sites accept, it would be great if also the batch system would be configured automatically • Here we are crossing the border between Deployment and Operation • An understanding has still be obtained on how to achieve that

  16. SUMMARY • The aim is still to obtain the model described last year • Uncertainties have been clarified • Close work with grid deployment to obtain a proper configuration of sites • Its clear that we have to move carefully • Certification testbed • Preproduction testbed • Production • Each step has to be followed up with actual tests of the functionality

More Related